Beat ChatGPT at Fortune-Telling—An Attempt to Optimize a Large Language Model
Steven Cheng, Mingyang Han, Zonghong Lu, W. L. Xu, Jingyan Yu · Applied and Computational Engineering · 2025
In the context of the growing popularity of fortune-telling in modern Chinese society, this paper attempts to construct and optimize a language model specifically for fortune-telling and compares its performance with mainstream models such as ChatGPT and Llama. Due to the antiquity and cultural particularity of Chinese fortune-telling systems such as Zhou Yi and Eight Characters, they face unique challenges. We generated a customized Chinese fortune-telling data set through ChatGPT, crawled the data on multiple websites, continuously optimized the instructions to build specialized GPTs, and improved its output by continuously optimizing the question design and design. We then evaluated and compared the model's responses in the fields of career, finance, and health, and the results showed that the optimized model outperformed ChatGPT in terms of accuracy and depth of prediction. The paper also highlights the importance of preserving Chinese cultural traditions through AI and explores the user-friendliness of GPT-based solutions when training custom models.